3
$\begingroup$

Is it possible to train a GAN model to inpaint an image taken from a specific setting(e.g. an office, woods, beach ...) after we have cropped out people of it?

For example, I used this repo's pretrained GAN model on Places2

I used a MaskRCNN model to segment people, and then cropped them out :

Original Image Image with people Cropped out

But when I fed the image to the mentioned GAN model, results weren't so great :

Inpainted Image

If it's possible, what is the best GAN architecture to use?

$\endgroup$
3
$\begingroup$

I would doubt there's a single correct answer for best available architecture, but the current best results come from this paper by NVIDIA and this technical report by Adobe. The latter paper is the second iteration of the repo you linked. Both papers focus on non-rectangular inpainting, which seems to be your targeted interest. Neither has publicly available code, however, so if you're interested in seeing what those models can accomplish you'll have to re-implement them yourself!

Edit: NVIDIA released a PyTorch implementation since the initial answer, here's the repo.

$\endgroup$
1
  • 1
    $\begingroup$ Thank a lot, that is in fact what I am looking for. Do you happen to know any good tutorials on implementing a GAN using TensorFlow based on a research paper like these? $\endgroup$ – Fakher Mokadem Jul 12 '18 at 8:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.